LibSVM turns all my training vectors into support vectors, why?

Question

I am trying to use SVM for News article classification.

I created a table that contains the features (unique words found in the documents) as rows. I created weight vectors mapping with these features. i.e if the article has a word that is part of the feature vector table that location is marked as 1 or else 0.

Ex:- Training sample generated...

1 1:1 2:1 3:1 4:1 5:1 6:1 7:1 8:1 9:1 10:1 11:1 12:1 13:1 14:1 15:1 16:1 17:1 18:1 19:1 20:1 21:1 22:1 23:1 24:1 25:1 26:1 27:1 28:1 29:1 30:1

As this is the first document all the features are present.

I am using 1, 0 as class labels.

I am using svm.Net for classification.

I gave 300 weight vectors manually classified as training data and the model generated is taking all the vectors as support vectors, which is surely overfitting.

My total features (unique words/row count in feature vector DB table) is 7610.

What could be the reason?

Because of this over fitting my project is now in pretty bad shape. It is classifying every article available as a positive article.

In LibSVM binary classification is there any restriction on the class label?

I am using 0, 1 instead of -1 and +1. Is that a problem?

Answer 1

You need to do some type of parameter search, also if the classes are unbalanced the classifier might get artificially high accuracies without doing much. This guide is good at teaching basic, practical things, you should probably read it